I am currently in the process of creating a simulation optimisation problem for a reservoir operation system. I will use 2 metaheuristic multi-objective optimisation algorithms(NSGA III is one) and I am trying to determine what is the best way to go about the decision variables. The standard approach is for the evaluation functions of a multi-objective problem to contain the vector of the decision variables , for example setting the monthly or daily flow releases or reservoir level data as decision variables. I have also seen approaches whereby the decision variables have been set based on parameterising the reservoir rule curve, therefore reducing the number of variables. However, the evaluation function does not directly contain the decision variables, as these would now be linked to the respective time series based on a specific parameterised relationship. Does anyone have any thoughts on what the best approach to adopt would be? I see that the latter is faster but heavily reliant on the parameterised relationship between the rule curve and levels/flow releases etc. Any help would be greatly appreciated.